lttng [OPTIONS] <COMMAND>
The LTTng project aims at providing highly efficient tracing tools for Linux. It's tracers help tracking down performance issues and debugging problems involving multiple concurrent processes and threads. Tracing across multiple systems is also possible.
The lttng command line tool from the lttng-tools package is used to control both kernel and user-space tracing. Every interactions with the tracer should be done by this tool or by the liblttng-ctl provided with the lttng-tools package.
LTTng uses a session daemon (lttng-sessiond(8)), acting as a tracing registry, which allows you to interact with multiple tracers (kernel and user-space) inside the same container, a tracing session. Traces can be gathered from the kernel and/or instrumented applications (lttng-ust(3)). Aggregating and reading those traces is done using the babeltrace(1) text viewer.
We introduce the notion of tracing domains which is essentially a type of tracer (kernel or user space for now). In the future, we could see a third tracer being for instance an hypervisor. For some commands, you'll need to specify on which domain the command applies (-u or -k). For instance, enabling a kernel event, you must specify the kernel domain to the command so we know for which tracer this event is for.
In order to trace the kernel, the session daemon needs to be running as root. LTTng provides the use of a tracing group (default: tracing). Whomever is in that group can interact with the root session daemon and thus trace the kernel. Session daemons can co-exist meaning that you can have a session daemon running as Alice that can be used to trace her applications along side with a root daemon or even a Bob daemon. We highly recommend to start the session daemon at boot time for stable and long term tracing.
Every user-space applications instrumented with lttng-ust(3), will automatically register to the session daemon. This feature gives you the ability to list available traceable applications and tracepoints on a per user basis. (See list command).
This program follow the usual GNU command line syntax with long options starting with two dashes. Below is a summary of the available options.
Add context to event(s) and/or channel(s). A context is basically extra information appended to a channel. For instance, you could ask the tracer to add the PID information for all events in a channel. You can also add performance monitoring unit counters (perf PMU) using the perf kernel API). For example, this command will add the context information 'prio' and two perf counters (hardware branch misses and cache misses), to all events in the trace data output: # lttng add-context -k -t prio -t perf:branch-misses -t perf:cache-misses Please take a look at the help (-h/--help) for a detailed list of available contexts. If no channel is given (-c), the context is added to all channels. Otherwise the context will be added only to the given channel (-c). If -s, --session is omitted, the session name is taken from the .lttngrc file.
OPTIONS:
-h, --help
Show summary of possible options and commands.
-s, --session NAME
Apply on session name.
-c, --channel NAME
Apply on channel name.
-k, --kernel
Apply for the kernel tracer
-u, --userspace
Apply for the user-space tracer
-t, --type TYPE
Context type. You can repeat this option on the command line. Please
use "lttng add-context -h" to list all available types.
Quantify LTTng overhead
The LTTng calibrate command can be used to find out the combined average
overhead of the LTTng tracer and the instrumentation mechanisms used. This
overhead can be calibrated in terms of time or using any of the PMU performance
counter available on the system.
For now, the only calibration implemented is that of the kernel function
instrumentation (kretprobes).
* Calibrate kernel function instrumentation
Let's use an example to show this calibration. We use an i7 processor with 4
general-purpose PMU registers. This information is available by issuing dmesg,
looking for "generic registers".
This sequence of commands will gather a trace executing a kretprobe hooked on
an empty function, gathering PMU counters LLC (Last Level Cache) misses
information (see lttng add-context --help to see the list of available PMU
counters).
# lttng create calibrate-function
# lttng enable-event calibrate --kernel --function lttng_calibrate_kretprobe
# lttng add-context --kernel -t perf:LLC-load-misses -t perf:LLC-store-misses \
-t perf:LLC-prefetch-misses
# lttng start
# for a in $(seq 1 10); do \
lttng calibrate --kernel --function;
done
# lttng destroy
# babeltrace $(ls -1drt ~/lttng-traces/calibrate-function-* | tail -n 1)
The output from babeltrace can be saved to a text file and opened in a
spreadsheet (e.g. oocalc) to focus on the per-PMU counter delta between
consecutive "calibrate_entry" and "calibrate_return" events. Note that these
counters are per-CPU, so scheduling events would need to be present to account
for migration between CPU. Therefore, for calibration purposes, only events
staying on the same CPU must be considered.
The average result, for the i7, on 10 samples:
Average Std.Dev.
perf_LLC_load_misses: 5.0 0.577
perf_LLC_store_misses: 1.6 0.516
perf_LLC_prefetch_misses: 9.0 14.742
As we can notice, the load and store misses are relatively stable across runs
(their standard deviation is relatively low) compared to the prefetch misses.
We can conclude from this information that LLC load and store misses can be
accounted for quite precisely, but prefetches within a function seems to behave
too erratically (not much causality link between the code executed and the CPU
prefetch activity) to be accounted for.
OPTIONS:
-h, --help
Show summary of possible options and commands.
-k, --kernel
Apply for the kernel tracer
-u, --userspace
Apply for the user-space tracer
--function
Dynamic function entry/return probe (default)
Create tracing session. A tracing session contains channel(s) which contains event(s). It is domain agnostic meaning that you can enable channels and events for either the user-space tracer and/or the kernel tracer. It acts like a container aggregating multiple tracing sources. On creation, a .lttngrc file is created in your $HOME directory containing the current session name. If NAME is omitted, a session name is automatically created having this form: 'auto-yyyymmdd-hhmmss'. If no -o, --output is specified, the traces will be written in $HOME/lttng-traces.
OPTIONS:
-h, --help
Show summary of possible options and commands.
--list-options
Simple listing of options
-o, --output PATH
Specify output path for traces
Using these options, each API call can be controlled individually. For
instance, -C does not enable the consumer automatically. You'll need the -e
option for that.
-U, --set-url=URL
Set URL for the consumer output destination. It is persistent for the
session lifetime. Redo the command to change it. This will set both
data and control URL for network.
-C, --ctrl-url=URL
Set control path URL. (Must use -D also)
-D, --data-url=URL
Set data path URL. (Must use -C also)
URL FORMAT:
proto://[HOST|IP][:PORT1[:PORT2]][/TRACE_PATH]
Supported protocols are (proto):
> file://...
Local filesystem full path.
> net://...
This will use the default network transport layer which is TCP for both
control (PORT1) and data port (PORT2). The default ports are
respectively 5342 and 5343. Note that net[6]:// is not yet supported.
> tcp[6]://...
Can only be used with -C and -D together
NOTE: IPv6 address MUST be enclosed in brackets '[]' (rfc2732)
EXAMPLES:
# lttng create -U net://192.168.1.42
Uses TCP and default ports for the given destination.
# lttng create -U net6://[fe80::f66d:4ff:fe53:d220]
Uses TCP, default ports and IPv6.
# lttng create s1 -U net://myhost.com:3229
Create session s1 and set its consumer to myhost.com on port 3229 for control.
Teardown tracing session Free memory on the session daemon and tracer side. It's gone! If NAME is omitted, the session name is taken from the .lttngrc file.
OPTIONS:
-h, --help
Show summary of possible options and commands.
-a, --all
Destroy all sessions
--list-options
Simple listing of options
Enable tracing channel To enable an event, you must enable both the event and the channel that contains it. If -s, --session is omitted, the session name is taken from the .lttngrc file. It is important to note that if a certain type of buffers is used, the session will be set with that type and all other subsequent channel need to have the same type.
OPTIONS:
-h, --help
Show this help
--list-options
Simple listing of options
-s, --session NAME
Apply on session name
-k, --kernel
Apply to the kernel tracer
-u, --userspace
Apply to the user-space tracer
--discard
Discard event when subbuffers are full (default)
--overwrite
Flight recorder mode : overwrites events when subbuffers are full
--subbuf-size SIZE
Subbuffer size in bytes {+k,+M,+G}
(default UST uid: 131072, UST pid: 4096, kernel: 262144, metadata: 4096)
Rounded up to the next power of 2.
--num-subbuf NUM
Number of subbuffers. (default UST uid: 4, UST pid: 4, kernel: 4, metadata: 2)
Rounded up to the next power of 2.
--switch-timer USEC
Switch subbuffer timer interval in µsec.
(default UST uid: 0, UST pid: 0, kernel: 0, metadata: 0)
--read-timer USEC
Read timer interval in µsec.
(default UST uid: 0, UST pid: 0, kernel: 200000, metadata: 0)
--output TYPE
Channel output type. Possible values: mmap, splice
(default UST uid: mmap, UST pid: mmap, kernel: splice, metadata: mmap)
--buffers-uid
Use per UID buffer (-u only). Buffers are shared between applications
that have the same UID.
--buffers-pid
Use per PID buffer (-u only). Each application has its own buffers.
--buffers-global
Use shared buffer for the whole system (-k only)
-C, --tracefile-size SIZE
Maximum size of each tracefile within a stream (in bytes).
0 means unlimited. (default: 0)
-W, --tracefile-count COUNT
Used in conjunction with -C option, this will limit the number
of files created to the specified count. 0 means unlimited. (default: 0)
EXAMPLES:
$ lttng enable-channel -C 4096 -W 32 chan1
For each stream, the maximum size of a trace file will be 4096 bytes divided
over a maximum of 32 different files. The file count is appended after
the stream number as seen in the following example. The last trace file is
smaller than 4096 since it was not completely filled.
~/lttng-traces/[...]/chan1_0_0 (4096)
~/lttng-traces/[...]/chan1_0_1 (4096)
~/lttng-traces/[...]/chan1_0_2 (3245)
~/lttng-traces/[...]/chan1_1_0 (4096)
...
$ lttng enable-channel -C 4096
This will create trace files of 4096 bytes and will create new ones as long as
there is data available.
Enable tracing event A tracing event is always assigned to a channel. If -c, --channel is omitted, a default channel named 'channel0' is created and the event is added to it. For the user-space tracer, using -a, --all is the same as using the wildcard "*". If -s, --session is omitted, the session name is taken from the .lttngrc file.
OPTIONS:
-h, --help
Show summary of possible options and commands.
--list-options
Simple listing of options
-s, --session NAME
Apply on session name
-c, --channel NAME
Apply on channel name
-a, --all
Enable all tracepoints and syscalls. This actually enable a single
wildcard event "*".
-k, --kernel
Apply for the kernel tracer
-u, --userspace
Apply for the user-space tracer
--tracepoint
Tracepoint event (default)
- userspace tracer supports wildcards at end of string. Don't forget to
quote to deal with bash expansion.
e.g.:
"*"
"app_component:na*"
--loglevel NAME
Tracepoint loglevel range from 0 to loglevel. Listed in the help (-h).
--loglevel-only NAME
Tracepoint loglevel (only this loglevel).
The loglevel or loglevel-only options should be combined with a
tracepoint name or tracepoint wildcard.
--probe [addr | symbol | symbol+offset]
Dynamic probe. Addr and offset can be octal (0NNN...), decimal (NNN...)
or hexadecimal (0xNNN...)
--function [addr | symbol | symbol+offset]
Dynamic function entry/return probe. Addr and offset can be octal
(0NNN...), decimal (NNN...) or hexadecimal (0xNNN...)
--syscall
System call event. Enabling syscalls tracing (kernel tracer), you will
not be able to disable them with disable-event. This is a known
limitation. You can disable the entire channel to do the trick.
--filter 'expression'
Set a filter on a newly enabled event. Filter expression on event
fields and context. Event recording depends on evaluation. Only
specify on first activation of a given event within a session.
Filter only allowed when enabling events within a session before
tracing is started. If the filter fails to link with the event
within the traced domain, the event will be discarded.
Currently, filter is only implemented for the user-space tracer.
Expression examples:
'intfield > 500 && intfield < 503'
'(stringfield == "test" || intfield != 10) && intfield > 33'
'doublefield > 1.1 && intfield < 5.3'
Wildcards are allowed at the end of strings:
'seqfield1 == "te*"'
In string literals, the escape character is a '\'. Use '\*' for
the '*' character, and '\\' for the '\' character. Wildcard
match any sequence of characters, including an empty sub-string
(match 0 or more characters).
Context information can be used for filtering. The examples
below show usage of context filtering on process name (with a
wildcard), process ID range, and unique thread ID for filtering.
The process and thread ID of running applications can be found
under columns "PID" and "LWP" of the "ps -eLf" command.
'$ctx.procname == "demo*"'
'$ctx.vpid >= 4433 && $ctx.vpid < 4455'
'$ctx.vtid == 1234'
Disable tracing channel Disabling a channel makes all event(s) in that channel to stop tracing. You can enable it back by calling lttng enable-channel NAME again. If -s, --session is omitted, the session name is taken from the .lttngrc file.
OPTIONS:
-h, --help
Show summary of possible options and commands.
--list-options
Simple listing of options
-s, --session NAME
Apply on session name
-k, --kernel
Apply for the kernel tracer
-u, --userspace
Apply for the user-space tracer
Disable tracing event The event, once disabled, can be re-enabled by calling lttng enable-event NAME again. If -s, --session is omitted, the session name is taken from the .lttngrc file.
OPTIONS:
-h, --help
Show summary of possible options and commands.
--list-options
Simple listing of options
-s, --session NAME
Apply on session name
-a, --all-events
Disable all events. This does NOT disable "*" but rather
every known events of the session.
-k, --kernel
Apply for the kernel tracer
-u, --userspace
Apply for the user-space tracer
List tracing session information.
With no arguments, it will list available tracing session(s).
With the session name, it will display the details of the session including
the trace file path, the associated channels and their state (activated
and deactivated), the activated events and more.
With -k alone, it will list all available kernel events (except the system
calls events).
With -u alone, it will list all available user-space events from registered
applications. Here is an example of 'lttng list -u':
PID: 7448 - Name: /tmp/lttng-ust/tests/hello/.libs/lt-hello
ust_tests_hello:tptest_sighandler (type: tracepoint)
ust_tests_hello:tptest (type: tracepoint)
You can now enable any event listed by using the name :
ust_tests_hello:tptest.
OPTIONS:
-h, --help
Show summary of possible options and commands.
--list-options
Simple listing of options
-k, --kernel
Select kernel domain
-u, --userspace
Select user-space domain.
SESSION OPTIONS:
-c, --channel NAME
List details of a channel
-d, --domain
List available domain(s)
Set current session name Will change the session name in the .lttngrc file.
OPTIONS:
-h, --help
Show summary of possible options and commands.
--list-options
Simple listing of options
Start tracing It will start tracing for all tracers for a specific tracing session. If NAME is omitted, the session name is taken from the .lttngrc file.
OPTIONS:
-h, --help
Show summary of possible options and commands.
--list-options
Simple listing of options
Stop tracing It will stop tracing for all tracers for a specific tracing session. Before returning, the command checks for data availability meaning that it will wait until the trace is readable for the session. Use --no-wait to avoid this behavior. If NAME is omitted, the session name is taken from the .lttngrc file.
OPTIONS:
-h, --help
Show summary of possible options and commands.
--list-options
Simple listing of options
--no-wait
Don't wait for data availability.
Show version information
OPTIONS:
-h, --help
Show summary of possible options and commands.
--list-options
Simple listing of options
View traces of a tracing session By default, the babeltrace viewer will be used for text viewing. If SESSION_NAME is omitted, the session name is taken from the .lttngrc file.
OPTIONS:
-h, --help
Show this help
--list-options
Simple listing of options
-t, --trace-path PATH
Trace directory path for the viewer
-e, --viewer CMD
Specify viewer and/or options to use
This will completely override the default viewers so
please make sure to specify the full command. The trace
directory path of the session will be appended at the end
to the arguments
Any other value above 10, please refer to <lttng/lttng-error.h> for a detailed list or use lttng_strerror() to get a human readable string of the error code.
Note that all command line options override environment variables.
If you encounter any issues or usability problem, please report it on our mailing list <lttng-dev@lists.lttng.org> to help improve this project or at https://bugs.lttng.org which is a bugtracker.
lttng is distributed under the GNU General Public License version 2. See the file COPYING for details.
A Web site is available at http://lttng.org for more information on the LTTng project.
You can also find our git tree at http://git.lttng.org.
Mailing lists for support and development: <lttng-dev@lists.lttng.org>.
You can find us on IRC server irc.oftc.net (OFTC) in #lttng.
Thanks to Yannick Brosseau without whom this project would never have been so lean and mean! Also thanks to the Ericsson teams working on tracing which helped us greatly with detailed bug reports and unusual test cases.
Thanks to our beloved packager Alexandre Montplaisir-Goncalves (Ubuntu and PPA maintainer) and Jon Bernard for our Debian packages.
Special thanks to Michel Dagenais and the DORSAL laboratory at Polytechnique de Montreal for the LTTng journey.
lttng-tools was originally written by Mathieu Desnoyers, Julien Desfossez and David Goulet. More people have since contributed to it. It is currently maintained by David Goulet <dgoulet@efficios.com>.